catalog_gwas_pliercanon_sep_ldclumping_r01d1000

Summary table

I used ‘qvalue’ R package to compute the fdr from p-values for each SNP and made a table to show the number of SNPs that pass the threshold. The thresholds are ‘fdr < 0.1’,‘fdr < 0.2’,‘pval < 5e-8’. The ‘num_significant_pairs’ indicates the number of significant pairs under each threshold. If a trait~factor pair has as least 1 significant SNP, we named it as ‘significant pair’.

number of significant trait-LV associations across all traits

	x
num_sigpairs_fdr010	146
num_sigpairs_fdr020	383

Info table

For each trait, I made a table to show the info of snps with fdr>0.2 in the factor ~ SNP + genotype pcs association test. For each trait,The LVs have more than one significant SNPs with FDR<0.2 are included.

The suffix ’_assoc’ here means that results are from factor ~ SNP + genotype pcs association test. The suffix ’_gwas’ here means results are from original GWAS results files. For EUR.CD, EUR.IBD, EUR.UC,T2D, asthma, the effectsize_gwas here means ‘ln(OR)’, for others, it means ‘beta’.

‘snp_ld’ here means the snps that in LD with the snp in each line.’ld_r2’ means the LD r-squared which is corresponding to the ‘snp_ld’ column. ‘cis-eqtl’ column indicates whether the snp is a cis-eqtl according to GTEx data. ‘cis_gene_hgnc’ and ‘cis_gene_hgnc’ is the genes that the snp influence when it act as cis-eqtl. ‘func’ and ‘func_gene’ are obtained from ANNOVAR, which indicating the snp function within the genes.

Enrichment analysis

For some promising trait-factor pairs (i.e. BMI-LV3, LV27, RBC-LV82, AsthmaLV68, WBC-LV119, Lymphocyte-LV23, LV78), I did enrichment analysis with WebGestalt. The analysis are under different settings:

1. ORA: For each factor, I sorted the genes that have non-zero loadings by their loadings, taking the top 25% as the gene set. The function database here are Reactome pathway; Disgenet + GLAD4U + OMIM disease dataset; geneontology biological process. The reference set affy hugene 2 0 st v1. Minimum number of genes for a category is 5, maximum number of genes for a category is 2000(default settings). All categories that have fdr<0.01 are listed.

BMI-LV27

BMI-LV76

BMI-LV90

PLT-LV49

RBC-LV82

Asthma-LV36

Asthma-LV39

WBC-LV6

Warning in instance$preRenderHook(instance): It seems your data is too big
for client-side DataTables. You may consider server-side processing: https://
rstudio.github.io/DT/server.html

WBC-LV119

Lymphocyte-LV23

Lymphocyte-LV26

LDL-LV125

WHR

Effect size plots & Resampling

For each trait, I made a plot of association with LV(indicating by beta in GWAS) vs association with trait(indicating by ln(odds ratio) or beta in GWAS) to show if the variants have the correlated effect direction. The effect sizes of Catalog GWAS and factor association tests are harmonized by TwoSampleMR R package to make the effect alleles in these two analysis identical. The LVs have more than one significant SNPs with FDR<0.2 are included in the plotting.Besides, for each plots, I fitted the points with intercept = 0. The pvalues and r-squared are shown on the plots.

I also relaxed the fdr threshold of the SNPs that used to make effect size plots(from 0.2 to 0.3/0.5)

For all pairs , I did resampling. I resampled the SNPs without replacement and fitted the points with intercept = 0 again and recorded the pvalues and r-squared. The resampling was repeated 1000 times.

I made a histogram to show the rsquared distribution from resampling. The red line in the plots are the rsquared in the origin analysis. The r_mean values in the plots are the mean values of rsquared in point fitting. The ‘p-value from resampling’ is computed by: (number of more extreme values)/(times of resampling).

Effectsize summary

To summarize the results above, I made a scatter plot of all trait-LV pairs. The x-axis shows the R2 in fitting, and the y-axis show the -log10-p values from resampling. Below are plots for three FDR cutoffs, 0.2, 0.3 and 0.5. To make the plot more informative, Each point represent a trait-LV pair. Different colors show the number of SNPs passing the FDR threshold for each pair.

Summary Table

Effect size plots - checking the reverse causality (19 traits)

To check if the effect size correlation is due to reverse causality: i.e. trait -> LV (trait causally affect LV), instead of LV -> trait (which is what we like to see). I used all SNPs associated with traits(pval<5E-8). The x-axis is the effects of these SNPs on trait, and y-axis is the effects on LV.

Some pair show p < 0.05, the result may be driven by the possible causal effect of LV -> trait. To test this, I removed the SNPs that are associated with LVs at FDR < 0.2 and made the plots again.

QQplots (19 traits)

Colocalization – 500kb (19 traits)

The colocalization analysis was performed using the approximate Bayes factor test implemented in the Coloc package. Coloc computes five posterior probabilities (PP0, PP1, PP2, PP3 and PP4), each corresponding to a hypothesis: H0, no association with either trait; H1, association with trait 1 but not with trait 2; H2, association with trait 2 but not with trait 1; H3, association with trait 1 and trait 2, two independent SNPs; H4, association with trait 1 and trait 2, one shared SNP. We ran Coloc with the default parameters and used PP4 to assess evidence of colocalization. We visualized the colocalization of factor - QTLs and GWAS associations using the LocusCompareR package.

SNP selection: 1. Chose the SNPs in the info table. 2. For each SNP, the region used in colocalization analysis is between [pos-100kb, pos+100kb]. 3. All SNPs in this region are included in alalysis.

Summary Table

Summary table

I used ‘qvalue’ R package to compute the fdr from p-values for each SNP and made a table to show the number of SNPs that pass the threshold. The thresholds are ‘fdr < 0.1’,‘fdr < 0.2’,‘pval < 5e-8’.

Info table

For each trait, I made a table to show the info of snps with fdr>0.2 in the factor ~ SNP + genotype pcs association test. For each trait,The LVs have more than one significant SNPs with FDR<0.2 are included.

The suffix ’_assoc’ here means that results are from factor ~ SNP + genotype pcs association test. The suffix ’_gwas’ here means results are from original GWAS results files. For EUR.CD, EUR.IBD, EUR.UC, the effectsize_gwas here means ‘ln(OR)’, for others, it means ‘beta’.

‘snp_ld’ here means the snps that in LD with the snp in each line.’ld_r2’ means the LD r-squared which is corresponding to the ‘snp_ld’ column. ‘cis-eqtl’ column indicates whether the snp is a cis-eqtl according to GTEx data. ‘cis_gene_hgnc’ and ‘cis_gene_hgnc’ is the genes that the snp influence when it act as cis-eqtl. ‘func’ and ‘func_gene’ are obtained from ANNOVAR, which indicating the snp function within the genes.

Enrichment analysis

For some promising trait-factor pairs (i.e. BMI-LV3, LV27, RBC-LV82, AsthmaLV68, WBC-LV119, Lymphocyte-LV23, LV78), I did enrichment analysis with WebGestalt. The analysis are under different settings:

1. ORA: For each factor, I sorted the genes in 1 by their loadings, taking the top 25% as the gene set. The function database here are Reactome pathway; Disgenet + GLAD4U + OMIM disease dataset; geneontology biological process. The reference set affy hugene 2 0 st v1. Minimum number of genes for a category is 5, maximum number of genes for a category is 2000(default settings). All categories that have fdr<0.01 are listed.

BMI-LV3

BMI-LV27

BMI-LV76

BMI-LV90

RBC-LV82

Asthma-LV68

WBC-LV119

Lymphocyte-LV23

Lymphocyte-LV78

2. GSEA: For each factor, I used all genes (zero loading genes included), taking the gene loadings as their gene score in GSEA. The function database is Reactome pathway.

BMI-LV3-Reactome BMI-LV3-GO BMI-LV3-phenotype BMI-LV3-disease-Disgenet

BMI-LV27-Reactome BMI-LV27-GO BMI-LV27-phenotype BMI-LV27-disease-Disgenet

BMI-LV76-Reactome BMI-LV76-GO BMI-LV76-phenotype BMI-LV76-disease-Disgenet

RBC-LV82-Reactome RBC-LV82-GO RBC-LV82-phenotype RBC-LV82-disease-Disgenet

Asthma-LV68-Reactome Asthma-LV68-GO Asthma-LV68-phenotype Asthma-LV68-disease-Disgenet

WBC-LV119-Reactome WBC-LV119-GO WBC-LV119-phenotype WBC-LV119-disease-Disgenet

Lymphocyte-LV23-Reactome Lymphocyte-LV23-GO Lymphocyte-LV23-phenotype Lymphocyte-LV23-disease-Disgenet

Lymphocyte-LV78 Lymphocyte-LV78-GO Lymphocyte-LV78-phenotype Lymphocyte-LV78-disease-Disgenet

3. GSEA: For each factor, I used all genes (zero loading genes included), taking the gene loadings as their gene score in GSEA. For those loadings = 0, I assigned a random number from normal distribution N(0,1e-5) to it to avoid ties. The function database is Reactome pathway.

BMI-LV3

BMI-LV27

BMI-LV76

RBC-LV82

Asthma-LV68

WBC-LV119

Lymphocyte-LV23

Lymphocyte-LV78

Effect size plots

For each trait, I made a plot of association with LV(indicating by beta in GWAS) vs association with trait(indicating by ln(odds ratio) or beta in GWAS) to show if the variants have the correlated effect direction. The effect sizes of Catalog GWAS and factor association tests are harmonized by TwoSampleMR R package to make the effect alleles in these two analysis identical. The LVs have more than one significant SNPs with FDR<0.2 are included in the plotting.Besides, for each plots, I fitted the points with intercept = 0. The pvalues and r-squared are shown on the plots.

effect size plots – more SNPs

For some promising trait-factor pairs (i.e. BMI-LV3, LV27,LV76, RBC-LV82, Asthma-LV68, WBC-LV119, Lymphocyte-LV23, LV78), I relaxed the fdr threshold of the SNPs that used to make effect size plots(from 0.2 to 0.3/0.5)

I also made a plot to show the the distribution of the SNPs’ fdr.

Effect size plots - checking the reverse causality

To check if the effect size correlation is due to reverse causality: i.e. trait -> LV (trait causally affect LV), instead of LV -> trait (which is what we like to see). I used all SNPs associated with traits(pval<5E-8). The x-axis is the effects of these SNPs on trait, and y-axis is the effects on LV.

Some pair show p < 0.05, the result may be driven by the possible causal effect of LV -> trait. To test this, I removed the SNPs that are associated with LVs at FDR < 0.2 and made the plots again.

Summary table

I used ‘qvalue’ R package to compute the fdr from p-values for each SNP and made a table to show the number of SNPs that pass the threshold. The thresholds are ‘fdr < 0.1’,‘fdr < 0.2’,‘pval < 5e-8’.

Info table

For each trait, I made a table to show the info of snps with fdr>0.2 in the factor ~ SNP + genotype pcs association test. For each trait,The LVs have more than one significant SNPs with FDR<0.2 are included.

The suffix ’_assoc’ here means that results are from factor ~ SNP + genotype pcs association test. The suffix ’_gwas’ here means results are from original GWAS results files. For EUR.CD, EUR.IBD, EUR.UC, the effectsize_gwas here means ‘ln(OR)’, for others, it means ‘beta’.

‘snp_ld’ here means the snps that in LD with the snp in each line.’ld_r2’ means the LD r-squared which is corresponding to the ‘snp_ld’ column. ‘cis-eqtl’ column indicates whether the snp is a cis-eqtl according to GTEx data. ‘cis_gene_hgnc’ and ‘cis_gene_hgnc’ is the genes that the snp influence when it act as cis-eqtl. ‘func’ and ‘func_gene’ are obtained from ANNOVAR, which indicating the snp function within the genes.

Effect size plots

For each trait, I made a plot of association with LV(indicating by beta in GWAS) vs association with trait(indicating by ln(odds ratio) or beta in GWAS) to show if the variants have the correlated effect direction. The effect sizes of Catalog GWAS and factor association tests are harmonized by TwoSampleMR R package to make the effect alleles in these two analysis identical. The LVs have more than one significant SNPs with FDR<0.2 are included in the plotting.Besides, for each plots, I fitted the points with intercept = 0. The pvalues and r-squared are shown on the plots.

Effect size plots- more SNPs

For some promising trait-factor pairs , I relaxed the fdr threshold of the SNPs that used to make effect size plots(from 0.2 to 0.3/0.5)

Effect size plots - checking the reverse causality

To check if the effect size correlation is due to reverse causality: i.e. trait -> LV (trait causally affect LV), instead of LV -> trait (which is what we like to see). I used all SNPs associated with traits(pval<5E-8). The x-axis is the effects of these SNPs on trait, and y-axis is the effects on LV.

Some pair show p < 0.05, the result may be driven by the possible causal effect of LV -> trait. To test this, I removed the SNPs that are associated with LVs at FDR < 0.2 and made the plots again.